Search bioRxiv⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.64898/2025.12.01.691347

Plant Face: Machine learning decodes genetic, environmental and developmental imprints in leaf appearance

Li, J.·Chen, Y.·Gao, Z.-h.·Zhou, X.·Xu, Z.·Zhang, Y.·Huang, K.·Ma, G.·Zhao, Y.-J.·Liu, X.·GUO, Y.
View original source ↗Export metadata ↓

Abstract

Leaf appearance is a crucial plant phenotype. However, traditional methods for extracting this information are inefficient, limiting its full utilization. Deep learning based on convolutional neural networks (CNNs) enables us to capture previously inaccessible information from images. In this study, we made the surprising discovery that the leaf appearance of each individual plant is unique. Using deep learning, leaves from one plant could be efficiently distinguished from those of another plant of the same species and cultivar. We term this phenomenon the "Plant Face" and suggest the potential to develop a "plant face recognition system," analogous to human facial recognition. We also applied similar methods to study the relationship between leaflet appearance and their position on compound leaves, leaf bilateral symmetry, and differences in leaves from twining stems with different chirality. These results collectively indicate that plant genetic characteristics, growth conditions, and developmental features can be stored within their appearance. With appropriate decoding, leaf appearance is poised to play an increasingly important role in phenomics. SignificanceThe saying "no two leaves in the world are identical" holds philosophical significance, as such variation encompasses considerable contingency and randomness. Here, we assert that no two trees have identical leaves; meaning that even for plants of the same species and cultivar, the leaf morphology of each individual plant is distinct at the population level, even though single leaves may overlap in appearance. Genetic, environmental, and developmental information is recorded in some manner within the phenotypic appearance of leaves. With advancements in computational technologies like artificial intelligence, this information can now be decoded. HighlightsO_LIThe leaves of each individual plant are statistically unique. C_LIO_LIThe relationship between leaflet appearances in compound leaves hints at their developmental patterns. C_LIO_LILeaves are not necessarily bilaterally symmetric in a statistical sense. C_LIO_LILeaves from stems with different chirality (twining direction) exhibit distinct appearances. C_LI

Explore related subjects

plant biologyPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Li, J., Chen, Y., Gao, Z.-h., Zhou, X., Xu, Z., Zhang, Y., Huang, K., Ma, G., Zhao, Y.-J., Liu, X., GUO, Y.. 2025-12-02. Plant Face: Machine learning decodes genetic, environmental and developmental imprints in leaf appearance. https://doi.org/10.64898/2025.12.01.691347

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

In-cell structural analysis reveals a distinctive chloroplast ribosome in Chlamydomonas reinhardtii

Chloroplast ribosomes synthesize plastid-encoded components of photosynthetic machinery, yet their structure and organization remain poorly understood. We combined cryo-focused ion beam milling, cryo-electron tomography and subtomogram averaging to determine native chloroplast ribosomes in Chlamydomonas reinhardtii. The 4.4-4.9 [A] structure revealed a large arch-like extension on the small subunit (SSU). Comparisons with bacterial and plant chloroplast ribosomes, supported by proteomics, AlphaFold3 predictions and a recent atomic model, indicate that the arch is formed by insertions and extensions in SSU proteins. Classification resolved active, thylakoid-associated ribosomes with density adjacent to the nascent peptide exit and an arch-moved state enriched among thylakoid-associated particles, with coordinated displacement of the arch and beak. Phylogenetic analysis revealed an evolutionary mosaic: the uS3c insertion is broadly distributed across Chlorophyceae, whereas the uS2c insertion, uS5c and PSRP7 are concentrated in Chlamydomonadales, with PSRP7 also in Sphaeropleales. Nuclear-encoded components were recruited stepwise onto a plastid-encoded scaffold, with all four under comparable purifying selection. These findings link a lineage-specific SSU extension to ribosome dynamics, thylakoid association and evolution, highlighting the value of in-cell structural analysis.

plant biology↗
bioRxiv2026

Implementation and calibration of the Vaganov-Shashkin model in the virtualRings R package

Process-based tree growth models provide a mechanistic framework for investigating how climate conditions regulate tree growth across daily to annual time scales. Yet, their broader application across species and environments is constrained by the limited accessibility in open-source environments and the difficulty of estimating physiological parameters that are rarely measured directly. Here, we present virtualRings, a new R package integrating the Vaganov-Shashkin model (VSM) and the RINGS3 models, and focus on the implementation and calibration of VSM. Using tree-ring width observations from seven Northern Hemisphere sites across various environmental conditions, we compared the traditional bootstrap-based calibration approach with the Covariance Matrix Adaptation Evolution Strategy (CMA-ES). CMA-ES improved agreement between simulated and observed radial tree growth and provided an efficient approach for model parameter estimation. We further evaluated practical CMA-ES settings to balance computational cost and performance and discussed its potential limitations. The virtualRings package provides an open and reproducible platform for tree growth simulation, facilitating the application of important process-based models across species and environments and the investigation of how temperature and moisture constraints regulate daily tree-ring formation across spatial and temporal scales.

plant biology↗
bioRxiv2026

Timing of transient darkness shapes carbon-nitrogen metabolism and sugar signaling in sugarcane

Fluctuating light is common in field environments. Yet, the mechanisms by which C4 crops coordinate carbon and nitrogen metabolism during short-term carbon deprivation remain poorly understood. Here, we imposed transient darkness at different phases of the diel cycle to assess how the timing of light loss affects photosynthesis, carbohydrate turnover, amino acid dynamics, and sugar-sensing pathways in commercial sugarcane leaves. Early-day darkness significantly impaired photosynthetic induction and revealed a temporal disconnect between stomatal and metabolic limitations, whereas midday and late-day treatments caused temporary, time-specific disruptions in carbon assimilation. These shifts altered the balance between sucrose preservation and catabolic mobilization, leading to treatment-dependent changes in starch reserves and free amino acids. Core circadian components largely maintained their phase relationships, but their amplitudes varied across treatments, consistent with partial decoupling from carbon status. Darkness also reorganized energy signaling, with SnRK1 and DIN6 responses associated with greater declines in sucrose. Notably, trehalose-pathway transcripts showed marked changes in network connectivity, with ScTPSIIG consistently emerging as a highly connected candidate associated with photosynthetic performance, water-use traits, sugar sensing, and amino acid metabolism. Overall, these results indicate that the timing of carbon limitation and residual sucrose availability shape distinct metabolic responses, while trehalose metabolism provides a candidate regulatory layer coordinating carbon-nitrogen adjustment during the diel cycle, highlighting class II TPS proteins as targets for functional investigation of metabolic resilience in sugarcane.

plant biology↗